TL;DR: Mistral released Voxtral 3, a family of transcribe and speech models it claims are state-of-the-art, bundled with a new voice agents API.
Summary: Mistral announced Voxtral 3, positioning it as state-of-the-art across both transcription and speech generation. The models are the engine behind Mistral's newly introduced voice agents API, which lets developers build spoken interfaces on top of Mistral's stack rather than stitching together separate STT and TTS vendors. The announcement came via Mistral's official X account with a demo video.
Why it matters: A single vendor covering both speech-to-text and speech output simplifies latency and integration work for anyone shipping voice agents. Worth testing the API against your current STT/TTS pipeline, especially on European-language and on-prem requirements where Mistral has an edge.
Source: x_com